WSClean version 2.9¶
Released 2020-03-27
WSClean 2.9 is released together with IDG version 0.7.
Summary of changes: WSClean+IDG is now much faster, in particular with fast-changing aterms. The combination of WSClean 2.9 + IDG 0.7 supports the efficient use of GPUs to grid large data volumes with complex a-terms. Furthermore, a new gridder that is more accurate compared to the default w-stacking gridder, and for some cases also faster! (in particular large images, many visibilities). The plan is also to move WSClean to Github in the next version of thereafter.
Major & visible changes:
Add new gridder “wgridder”, a stand-alone gridder written by Martin Reinecke that is now integrated in WSClean.
Applying a second aterm to all antennas is possible by setting dimension of antenna axis to one in aterm fits file.
Make source list output possible in IDG.
The FFTs are now performed with floats instead of doubles, thereby descreasing memory usage by 2 and increasing the speed anywhere from 20-100%.
LOFAR image-based beam correction (i.e. no aterm correction) now uses full Mueller matrix calculations instead of Jones, which removes some biases in beam correction that were seen.
New options
-reuse-psf
andreuse-dirty
to skip these steps when they already exist.WSClean makes use of IDG’s API changes to allocate optimal buffer sizes, greatly increasing the speed of IDG, and solving allocation problems on low-mem platforms
When parallel deconvolution is used, the logging is now much improved: only one thread reports its full progress.
Correct smallest multi-scale scale when using Gaussian tapering. Multi-scale cleaning was using just the longest baseline for determining the smallest scale to clean with. This causes tapered imaging to almost never use the smallest scale.
Add a ‘dry run’ mode, requested by J. Morgan. Fixes #162.
Bug fixes & other changes:
Stokes V imaging with R/L pols results in zero image (reported by Chiara Stuardi)
Don’t allow direct MS access for multi spw sets, to prevent a deadlock.
Output error when trying to combine parallelgridding with IDG.
Don’t incorrectly apply the average beam when using idg + apply-primary-beam.
Support newer casacore versions.
Remove dependency on boost thread.
Image the gridding kernel as option was removed, as both IDG and WGridder don’t support it.
Fix beam calculation for 1 timestep mses
Fix a crash when using pb correction + deconvolution channels
Solve reported problem with HDF5/Intel compilation